Skip to content

[DA] Add check for base pointer invariance #148241

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jul 25, 2025
Merged

[DA] Add check for base pointer invariance #148241

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4977f6f
[DA] Add check for base pointer invariance
kasuga-fj Jul 11, 2025
1df731b
Add test
kasuga-fj Jul 17, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 17 additions & 0 deletions llvm/lib/Analysis/DependenceAnalysis.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3383,6 +3383,10 @@ bool DependenceInfo::tryDelinearize(Instruction *Src, Instruction *Dst,
SrcSubscripts, DstSubscripts))
return false;

assert(isLoopInvariant(SrcBase, SrcLoop) &&
isLoopInvariant(DstBase, DstLoop) &&
"Expected SrcBase and DstBase to be loop invariant");

int Size = SrcSubscripts.size();
LLVM_DEBUG({
dbgs() << "\nSrcSubscripts: ";
Expand Down Expand Up @@ -3666,6 +3670,19 @@ DependenceInfo::depends(Instruction *Src, Instruction *Dst,
SCEVUnionPredicate(Assume, *SE));
}

// Even if the base pointers are the same, they may not be loop-invariant. It
// could lead to incorrect results, as we're analyzing loop-carried
// dependencies. Src and Dst can be in different loops, so we need to check
// the base pointer is invariant in both loops.
Loop *SrcLoop = LI->getLoopFor(Src->getParent());
Loop *DstLoop = LI->getLoopFor(Dst->getParent());
if (!isLoopInvariant(SrcBase, SrcLoop) ||
!isLoopInvariant(DstBase, DstLoop)) {
LLVM_DEBUG(dbgs() << "The base pointer is not loop invariant.\n");
return std::make_unique<Dependence>(Src, Dst,
SCEVUnionPredicate(Assume, *SE));
}

uint64_t EltSize = SrcLoc.Size.toRaw();
const SCEV *SrcEv = SE->getMinusSCEV(SrcSCEV, SrcBase);
const SCEV *DstEv = SE->getMinusSCEV(DstSCEV, DstBase);
Expand Down
6 changes: 3 additions & 3 deletions llvm/test/Analysis/DependenceAnalysis/Banerjee.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -113,7 +113,7 @@ define void @banerjee1(ptr %A, ptr %B, i64 %m, i64 %n) nounwind uwtable ssp {
; CHECK-NEXT: Src: %2 = load i64, ptr %arrayidx6, align 8 --> Dst: store i64 %2, ptr %B.addr.12, align 8
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store i64 %2, ptr %B.addr.12, align 8 --> Dst: store i64 %2, ptr %B.addr.12, align 8
; CHECK-NEXT: da analyze - output [* *]!
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All the test changes except for FlipFlopBaseAddress.ll are related to dependencies between memory accesses of the form *B++.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

*B++ should be fine though, right? Should the condition possibly be loop invariant or addrec for the appropriate loop?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

*B++ should be fine though, right?

Yes, I think.

Should the condition possibly be loop invariant or addrec for the appropriate loop?

Probably yes, but I'm not quite sure what qualifies as an "appropriate loop". In this case, the SCEV value for %B.addr.12 is {%B.addr.06,+,8}<nuw><%for.body3> rather than something like {{%B,+,(8 * %m)}<%for.cond1.preheader>,+,8}<%for.body3>. Can we infer anything from the former SCEV representation?

; CHECK-NEXT: da analyze - confused!
;
; NORMALIZE-LABEL: 'banerjee1'
; NORMALIZE-NEXT: Src: store i64 0, ptr %arrayidx, align 8 --> Dst: store i64 0, ptr %arrayidx, align 8
Expand All @@ -127,7 +127,7 @@ define void @banerjee1(ptr %A, ptr %B, i64 %m, i64 %n) nounwind uwtable ssp {
; NORMALIZE-NEXT: Src: %2 = load i64, ptr %arrayidx6, align 8 --> Dst: store i64 %2, ptr %B.addr.12, align 8
; NORMALIZE-NEXT: da analyze - confused!
; NORMALIZE-NEXT: Src: store i64 %2, ptr %B.addr.12, align 8 --> Dst: store i64 %2, ptr %B.addr.12, align 8
; NORMALIZE-NEXT: da analyze - output [* *]!
; NORMALIZE-NEXT: da analyze - confused!
;
; DELIN-LABEL: 'banerjee1'
; DELIN-NEXT: Src: store i64 0, ptr %arrayidx, align 8 --> Dst: store i64 0, ptr %arrayidx, align 8
Expand All @@ -141,7 +141,7 @@ define void @banerjee1(ptr %A, ptr %B, i64 %m, i64 %n) nounwind uwtable ssp {
; DELIN-NEXT: Src: %2 = load i64, ptr %arrayidx6, align 8 --> Dst: store i64 %2, ptr %B.addr.12, align 8
; DELIN-NEXT: da analyze - confused!
; DELIN-NEXT: Src: store i64 %2, ptr %B.addr.12, align 8 --> Dst: store i64 %2, ptr %B.addr.12, align 8
; DELIN-NEXT: da analyze - output [* *]!
; DELIN-NEXT: da analyze - confused!
;
entry:
%cmp4 = icmp sgt i64 %n, 0
Expand Down
79 changes: 67 additions & 12 deletions llvm/test/Analysis/DependenceAnalysis/FlipFlopBaseAddress.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,11 +8,11 @@
define float @bug41488_test1(float %f) {
; CHECK-LABEL: 'bug41488_test1'
; CHECK-NEXT: Src: %0 = load float, ptr %p, align 4 --> Dst: %0 = load float, ptr %p, align 4
; CHECK-NEXT: da analyze - input [*]!
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: %0 = load float, ptr %p, align 4 --> Dst: store float %f, ptr %q, align 4
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store float %f, ptr %q, align 4 --> Dst: store float %f, ptr %q, align 4
; CHECK-NEXT: da analyze - output [*]!
; CHECK-NEXT: da analyze - confused!
;
entry:
%g = alloca float, align 4
Expand All @@ -34,11 +34,11 @@ for.cond.cleanup:
define void @bug41488_test2(i32 %n) {
; CHECK-LABEL: 'bug41488_test2'
; CHECK-NEXT: Src: %0 = load float, ptr %p, align 4 --> Dst: %0 = load float, ptr %p, align 4
; CHECK-NEXT: da analyze - input [*]!
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: %0 = load float, ptr %p, align 4 --> Dst: store float 0.000000e+00, ptr %q, align 4
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store float 0.000000e+00, ptr %q, align 4 --> Dst: store float 0.000000e+00, ptr %q, align 4
; CHECK-NEXT: da analyze - output [*]!
; CHECK-NEXT: da analyze - confused!
;
entry:
%g = alloca float, align 4
Expand Down Expand Up @@ -68,7 +68,7 @@ define void @bug53942_foo(i32 noundef %n, ptr noalias nocapture noundef writeonl
; CHECK-NEXT: Src: %.pre = load double, ptr %B, align 8 --> Dst: store double %.pre, ptr %arrayidx2, align 8
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store double %.pre, ptr %arrayidx2, align 8 --> Dst: store double %.pre, ptr %arrayidx2, align 8
; CHECK-NEXT: da analyze - output [*]!
; CHECK-NEXT: da analyze - confused!
;
entry:
%cmp8 = icmp sgt i32 %n, 1
Expand Down Expand Up @@ -99,11 +99,11 @@ for.body: ; preds = %for.body.preheader,
define void @bug53942_bar(i32 noundef %n, ptr noalias noundef %A, ptr noalias noundef %B) {
; CHECK-LABEL: 'bug53942_bar'
; CHECK-NEXT: Src: %0 = load double, ptr %arrayidx, align 8 --> Dst: %0 = load double, ptr %arrayidx, align 8
; CHECK-NEXT: da analyze - input [*]!
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: %0 = load double, ptr %arrayidx, align 8 --> Dst: store double %0, ptr %arrayidx8, align 8
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store double %0, ptr %arrayidx8, align 8 --> Dst: store double %0, ptr %arrayidx8, align 8
; CHECK-NEXT: da analyze - output [*]!
; CHECK-NEXT: da analyze - confused!
;
entry:
br label %for.cond
Expand Down Expand Up @@ -166,14 +166,14 @@ for.end: ; preds = %for.cond.cleanup
; (j % 2 == 0 ? A[i][j] : A[i][j+1]) = 1;
; }
;
; FIXME: There are loop-carried dependencies between the store instruction. For
; There are loop-carried dependencies between the store instruction. For
; example, the value of %ptr0 when (i, j) = (0, 1) is %A+8, which is the same
; as when (i, j) = (0, 2).

define void @non_invariant_baseptr_with_identical_obj(ptr %A) {
; CHECK-LABEL: 'non_invariant_baseptr_with_identical_obj'
; CHECK-NEXT: Src: store i32 1, ptr %idx, align 4 --> Dst: store i32 1, ptr %idx, align 4
; CHECK-NEXT: da analyze - none!
; CHECK-NEXT: da analyze - confused!
;
entry:
br label %loop.i.header
Expand Down Expand Up @@ -216,13 +216,13 @@ exit:
; Similar to the above case, but ptr0 is loop-invariant with respsect to the
; k-loop.
;
; FIXME: Same as the above case, there are loop-carried dependencies between
; the store.
; Same as the above case, there are loop-carried dependencies between the
; store.

define void @non_invariant_baseptr_with_identical_obj2(ptr %A) {
; CHECK-LABEL: 'non_invariant_baseptr_with_identical_obj2'
; CHECK-NEXT: Src: store i32 1, ptr %idx, align 4 --> Dst: store i32 1, ptr %idx, align 4
; CHECK-NEXT: da analyze - none!
; CHECK-NEXT: da analyze - confused!
;
entry:
br label %loop.i.header
Expand Down Expand Up @@ -259,3 +259,58 @@ loop.i.latch:
exit:
ret void
}

; Pseudo-code that is approximately semantically equivalent to the below IR:
;
; void f(int A[][32]) {
; for (int i = 0; i < 100; i++)
; for (int j = 0; j < 15; j++) {
; int offset = (j % 2 == 0) ? 1 : 0;
; A[i][2 * j + offset + 0] = 1;
; A[i][2 * j + offset + 1] = 1;
; }
; }
;
; There are loop-carried dependencies between the two stores. For example,
; A[0][2] is accessed from both the former one when (i, j) = (0, 1) and the
; latter one when (i, j) = (0, 0).
;
define void @non_invariant_baseptr_with_identical_obj3(ptr %A) {
; CHECK-LABEL: 'non_invariant_baseptr_with_identical_obj3'
; CHECK-NEXT: Src: store i32 1, ptr %idx0, align 4 --> Dst: store i32 1, ptr %idx0, align 4
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store i32 1, ptr %idx0, align 4 --> Dst: store i32 1, ptr %idx1, align 4
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store i32 1, ptr %idx1, align 4 --> Dst: store i32 1, ptr %idx1, align 4
; CHECK-NEXT: da analyze - confused!
;
entry:
br label %loop.i.header

loop.i.header:
%i = phi i32 [ 0, %entry ], [ %i.inc, %loop.i.latch ]
%A1 = getelementptr i32, ptr %A, i32 1
br label %loop.j

loop.j:
%j = phi i32 [ 0, %loop.i.header ], [ %j.inc, %loop.j ]
%ptr0 = phi ptr [ %A1, %loop.i.header ], [ %ptr1, %loop.j ]
%ptr1 = phi ptr [ %A, %loop.i.header ], [ %ptr0, %loop.j ]
%j2_0 = shl i32 %j, 1
%j2_1 = add i32 %j2_0, 1
%idx0 = getelementptr [32 x i32], ptr %ptr0, i32 %i, i32 %j2_0
%idx1 = getelementptr [32 x i32], ptr %ptr0, i32 %i, i32 %j2_1
store i32 1, ptr %idx0
store i32 1, ptr %idx1
%j.inc = add i32 %j, 1
%cmp.j = icmp slt i32 %j.inc, 15
br i1 %cmp.j, label %loop.j, label %loop.i.latch

loop.i.latch:
%i.inc = add i32 %i, 1
%cmp.i = icmp slt i32 %i.inc, 100
br i1 %cmp.i, label %loop.i.header, label %exit

exit:
ret void
}
8 changes: 4 additions & 4 deletions llvm/test/Analysis/DependenceAnalysis/GCD.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -398,7 +398,7 @@ define void @gcd6(i64 %n, ptr %A, ptr %B) nounwind uwtable ssp {
; CHECK-NEXT: Src: %2 = load i32, ptr %arrayidx9, align 4 --> Dst: store i32 %2, ptr %B.addr.12, align 4
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store i32 %2, ptr %B.addr.12, align 4 --> Dst: store i32 %2, ptr %B.addr.12, align 4
; CHECK-NEXT: da analyze - output [* *]!
; CHECK-NEXT: da analyze - confused!
;
entry:
%cmp4 = icmp sgt i64 %n, 0
Expand Down Expand Up @@ -475,7 +475,7 @@ define void @gcd7(i32 %n, ptr %A, ptr %B) nounwind uwtable ssp {
; CHECK-NEXT: Src: %11 = load i32, ptr %arrayidx12, align 4 --> Dst: store i32 %11, ptr %B.addr.12, align 4
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store i32 %11, ptr %B.addr.12, align 4 --> Dst: store i32 %11, ptr %B.addr.12, align 4
; CHECK-NEXT: da analyze - output [* *]!
; CHECK-NEXT: da analyze - confused!
;
entry:
%0 = zext i32 %n to i64
Expand Down Expand Up @@ -566,7 +566,7 @@ define void @gcd8(i32 %n, ptr %A, ptr %B) nounwind uwtable ssp {
; CHECK-NEXT: Src: %5 = load i32, ptr %arrayidx12, align 4 --> Dst: store i32 %5, ptr %B.addr.12, align 4
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store i32 %5, ptr %B.addr.12, align 4 --> Dst: store i32 %5, ptr %B.addr.12, align 4
; CHECK-NEXT: da analyze - output [* *]!
; CHECK-NEXT: da analyze - confused!
;
entry:
%cmp4 = icmp sgt i32 %n, 0
Expand Down Expand Up @@ -650,7 +650,7 @@ define void @gcd9(i32 %n, ptr %A, ptr %B) nounwind uwtable ssp {
; CHECK-NEXT: Src: %11 = load i32, ptr %arrayidx12, align 4 --> Dst: store i32 %11, ptr %B.addr.12, align 4
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store i32 %11, ptr %B.addr.12, align 4 --> Dst: store i32 %11, ptr %B.addr.12, align 4
; CHECK-NEXT: da analyze - output [* *]!
; CHECK-NEXT: da analyze - confused!
;
entry:
%0 = zext i32 %n to i64
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/Analysis/DependenceAnalysis/NonAffineExpr.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@ define void @f(ptr %a, i32 %n, i1 %arg) align 2 {
; CHECK-NEXT: Src: %t.2 = load ptr, ptr %a, align 4 --> Dst: %t.4 = load i32, ptr %t.3, align 4
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: %t.4 = load i32, ptr %t.3, align 4 --> Dst: %t.4 = load i32, ptr %t.3, align 4
; CHECK-NEXT: da analyze - input [* *]!
; CHECK-NEXT: da analyze - confused!
;
for.preheader:
%t.0 = ashr exact i32 %n, 3
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/Analysis/DependenceAnalysis/Preliminary.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,7 +69,7 @@ define void @p2(i64 %n, ptr %A, ptr %B) nounwind uwtable ssp {
; CHECK-NEXT: Src: %0 = load i64, ptr %arrayidx17, align 8 --> Dst: store i64 %0, ptr %B.addr.24, align 8
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store i64 %0, ptr %B.addr.24, align 8 --> Dst: store i64 %0, ptr %B.addr.24, align 8
; CHECK-NEXT: da analyze - output [* * *]!
; CHECK-NEXT: da analyze - confused!
;
entry:
%cmp10 = icmp sgt i64 %n, 0
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/Analysis/DependenceAnalysis/PreliminaryNoValidityCheckFixedSize.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,7 @@ define void @p2(i64 %n, ptr %A, ptr %B) nounwind uwtable ssp {
; CHECK-NEXT: Src: %0 = load i64, ptr %arrayidx17, align 8 --> Dst: store i64 %0, ptr %B.addr.24, align 8
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store i64 %0, ptr %B.addr.24, align 8 --> Dst: store i64 %0, ptr %B.addr.24, align 8
; CHECK-NEXT: da analyze - output [* * *]!
; CHECK-NEXT: da analyze - confused!
;
; LIN-LABEL: 'p2'
; LIN-NEXT: Src: store i64 %i.011, ptr %arrayidx8, align 8 --> Dst: store i64 %i.011, ptr %arrayidx8, align 8
Expand All @@ -42,7 +42,7 @@ define void @p2(i64 %n, ptr %A, ptr %B) nounwind uwtable ssp {
; LIN-NEXT: Src: %0 = load i64, ptr %arrayidx17, align 8 --> Dst: store i64 %0, ptr %B.addr.24, align 8
; LIN-NEXT: da analyze - confused!
; LIN-NEXT: Src: store i64 %0, ptr %B.addr.24, align 8 --> Dst: store i64 %0, ptr %B.addr.24, align 8
; LIN-NEXT: da analyze - output [* * *]!
; LIN-NEXT: da analyze - confused!
;
entry:
%cmp10 = icmp sgt i64 %n, 0
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/Analysis/DependenceAnalysis/SymbolicRDIV.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -437,7 +437,7 @@ define void @symbolicrdiv6(ptr %A, ptr %B, i64 %n1, i64 %n2) nounwind uwtable ss
; CHECK-NEXT: Src: %0 = load i32, ptr %arrayidx4, align 4 --> Dst: store i32 %0, ptr %B.addr.12, align 4
; CHECK-NEXT: da analyze - confused!
; CHECK-NEXT: Src: store i32 %0, ptr %B.addr.12, align 4 --> Dst: store i32 %0, ptr %B.addr.12, align 4
; CHECK-NEXT: da analyze - output [* *]!
; CHECK-NEXT: da analyze - confused!
;
entry:
%cmp4 = icmp eq i64 %n1, 0
Expand Down
Loading